2.1 Attributing intention

One claim of this work is that meaning based in sociophonetics is generally less likely to be taken to be intended by the speaker than that based in the semantics of morphosyntactic objects. This raises the general question, at the center of this work, of when a potential consequence of an action will be viewed as intended.

To address this question, I begin with the notion of an accessible alternative, which I characterize in (2).

In essence, an alternative was accessible for agent $ A $ if and only if it was an action that $ A $ could have performed and was on some level on $ A $ ’s radar.

Of course, just because an alternative is accessible does not mean it is desirable. Actions and their alternatives can come with various potential costs, which may be realized in physical, financial, social, or other terms. Certain actions, for instance, require great effort, making them generally less attractive than other actions. Alternatives also carry potential benefits, which we may think of in terms of the likelihood with which they will bring about desirable outcomes. Such potential costs and benefits determine how attractive a given accessible alternative is to an agent and, as I will discuss shortly, they in turn play an important role in assessing whether a consequence of an action was intended.

I now turn to a principle governing whether an observer takes a consequence of an action to have been intended, given in (3). The principle revolves around how favorably the observer thinks the agent would view the relevant outcome, how likely the observer thinks the agent would think the action was to effect that outcome, and how the observed action relates to alternatives that appear to have been more likely or less likely to effect that outcome.

Underlying (3) and the ensuing discussion is the presumption that agents and their observers are rational in the sense of attempting to maximize the net benefits (benefits less costs) of what they do (cf. Horn Reference Horn and Ward2004; Wilson & Sperber Reference Wilson and Sperber2004), and that this presumption of rationality is common ground among them. Where rationality of this kind is not assumed, the dynamics outlined below fall apart. It is also worth emphasizing that I do not intend to make any assumptions about the degree of consciousness under which (3) operates.

The ‘desirability’ factor in (3a) is perhaps obvious. We are more likely to think $ A $ intended a particular outcome for $ A $ ’s action the more we think that $ A $ would desire that outcome. If a footballer misses a shot on goal in an important match, we are generally unlikely to think the miss was intended; typically, players want not to miss. On the contrary, if we later learn that the player had bet a large sum on the match against their own team, we have good reason to think the player would want to miss the goal and thus that they intended to do so. The ‘expectation’ factor in (3b) is similarly straightforward. The less $ A $ believes $ \alpha $ would bring about $ c $ , the less sense there is in $ A $ attempting to effect $ c $ via $ \alpha $ . If, for instance, a two-year-old pulls a fire alarm and we doubt that the child was aware of the likely consequences of doing so, we likewise have reason to doubt the child intended to clear the building of people, etc. But if an adult who presumably has full knowledge of the purpose of fire alarms does so, there is good reason to think that that person indeed intended to clear the building.

Examples (3c) and (3d) are more complex because they involve not just the action itself but also how it relates to alternative actions. Example (3c), the ‘ $ M $ -alternatives’ factor, says that we are more likely to believe that $ A $ intended some outcome $ c $ the less we think $ A $ thought $ A $ was forgoing (in net benefits) by selecting $ \alpha $ over alternatives that were apparently accessible and more likely than $ \alpha $ to effect $ c $ . As an example, imagine we learn that an individual, Alex, said the word damn somewhat frustratedly, and we want to know whether Alex intended to express extreme displeasure, as opposed to, say, only moderate displeasure. Assume that Alex has at other times, especially when very upset, used a more taboo expletive than damn to express displeasure. Thinking in terms of (3c), our $ c $ here is expressing extreme displeasure, and a potential $ M $ -alternative – that is, an alternative to saying ‘damn’ that we think Alex would regard as accessible and more likely to express extreme displeasure – would be saying that stronger expletive instead.

Now let’s consider two possible settings for Alex’s utterance. In the first, Alex is among friends who frequently swear. In the second (holding other factors fixed – e.g. Alex’s delivery), Alex is at school teaching a class of kindergartners, where swearing is very much frowned upon. Intuitively speaking, we have greater reason to think Alex is very upset from Alex’s saying damn in front of kindergartners than among friends. In the kindergarten scenario, Alex, even if deeply upset, has good reason to avoid a stronger expletive (e.g. the potential of censure), whereas in the other scenario there is no great social cost to using a stronger expletive. Putting things again in terms of (3c), we could say that uttering the stronger expletive offers less in net benefits in the kindergarten scenario than in the friends scenario, because in the kindergarten scenario uttering a stronger swear comes at a higher cost. Thus, in accordance with (3c), we are more likely to think Alex intended to express extreme displeasure by saying damn in the kindergarten scenario, where the net benefits of using a stronger expletive are less.

A related logic underlies the ‘ $ L $ -alternatives’ factor spelled out in (3d), which says that we are more likely to think $ A $ intended to effect $ c $ the more we think $ A $ would think alternatives less likely to effect $ c $ had to offer relative to $ \alpha $ . To illustrate, consider the hypothetical case of Pat, who is certain to buy one of two houses that are identical except that one is on the east side of the street and the other is directly across from it on the west. Suppose Pat buys the house on the east side, and we are curious as to whether Pat wanted a house on that side in particular. Here, then, our $ c $ is having Pat’s home be on the east side of the street and the single $ L $ -alternative – in this case, an alternative action we believe Pat would think was less likely to effect $ c $ – is buying the house on the west side.

Again, comparing two versions of this scenario will illustrate the substance of (3d). In the first, suppose that the house on the west side cost $25,000 more than the other. In that case, it is not clear that Pat wanted to live on the east side in particular – perhaps the lower price was the reason for selecting that house. Now suppose the house on the west side cost $25,000 less than the other. In that case, we have greater reason to think Pat wanted to live on the east side in particular, because Pat could have paid considerably less for an identical house on the other side. Put in terms of (3d), the net benefits of opting for the $ L $ -alternative – namely, buying the house on the west side of the street – are higher in the second case, where doing so would cost $25,000 less than Pat’s actual action, than in the first, where it would cost $25,000 more. Accordingly, as (3d) suggests, we are more likely to think Pat intended to have a home on the east side of the street in particular in the second scenario, where the house on the west side costs less.

As we transition to thinking more specifically about intention in communication, it is worth noting that this $ L $ -alternatives factor provides grounds for doubting that Pooh’s bees intended to communicate something by their buzzing, or even specifically intended to issue a buzzing sound. Assuming the bees wished to move themselves swiftly, any alternative action less likely to involve buzzing (e.g. walking, waiting to be carried by the wind, etc.) would have little to offer. Hence, according to (3d), the buzzing was quite plausibly not a goal in itself but rather incidental to a goal of locomotion. This is akin to Pat buying a house on the east side of the street when it is $25,000 cheaper than the alternative – for all we know, it was saving money, not living on the east side in particular, that led to Pat’s purchase of that house. The dynamics involving $ L $ -alternatives in intention-attribution will play a prominent role in the material to come.

To summarize, (3) relates the likelihood of ascribing to an agent $ A $ the intention to effect some outcome $ c $ via action $ \alpha $ to four factors: (i) $ c $ ’s apparent desirability; (ii) how much we think $ A $ would expect $ \alpha $ to effect $ c $ ; and $ \alpha $ ’s apparent attractiveness relative to alternative actions that seem (iii) more or (iv) less likely to effect $ c $ . Thus far, I have discussed actions and outcomes in very broad terms, with examples from multiple domains, to illustrate this principle’s generality. Let us now focus more narrowly on a particular kind of action: utterances.

2.2 Intention and language meaning, broadly construed

Like actions in general, utterances have potentially many different parts, costs, benefits, and consequences. Among an utterance’s possible consequences are effects on a hearer. An utterance may for instance cause a hearer to feel a certain way, recall certain concepts, or change their beliefs. In my view, meaning studies should ultimately encompass a wide range of impacts that utterances have on hearers’ emotional and mental states, but I will focus on belief states here.

In talking about meaning and the way the term meaning is used, it will be useful to have a broad working notion of meaning that is general enough to accommodate the diversity of phenomena-labeled meaning in the literature. The notion described here is not meant to be absolute; in other contexts, other notions may be warranted. For our purposes, we might think of a form’s meaning in the abstract as its potential contribution to suggesting something about the world to a given hearer in a given context. This accords with the idea that a word’s semantic meaning is a function that may be composed with other word meanings to form a proposition (e.g. Montague Reference Montague and Visentini1970), which, when issued via an utterance, may suggest something about the world. It likewise accords with the idea that the meaning of an abstract variant of a sociolinguistic variable is an indexical field (Eckert Reference Eckert2008) – a collection of indexically associated stances, traits, etc. – which, when used, may suggest that the speaker embodies or wishes to mark as relevant some subset of that field. Moving away from forms in the abstract, we might say that for a given hearer, the meaning, in this very broad sense, of a particular instantiation of a form in practice is whatever contribution it actually makes to suggesting something to that hearer.

This very general notion of meaning-in-practice is stated relative to the hearer and says nothing about intention. Let’s now incorporate the perspective of the speaker. Suppose a speaker $ S $ ’s utterance $ u $ suggests something $ p $ to a hearer $ H $ . Example (4) enumerates three relations that may obtain between $ S $ , $ H $ , $ u $ , and $ p $ of particular interest herein, related to Grice’s (Reference Grice1957) observations.

As alluded to in Section 1, Grice (Reference Grice1957) distinguishes between two types of meaning: ‘natural’ and ‘nonnatural’. Natural meaning roughly coincides with (4a), where something provides information without any particular intention behind it. (Grice’s discussion suggests something a bit stricter, but this suffices for our purposes.) This is the ‘Those spots mean measles’ case. In contrast, to say that $ S $ meant something by some utterance $ u $ in the ‘nonnatural’ sense, according to Grice (Reference Grice1957: 385), ‘is (roughly) equivalent to “[ $ S $ ] intended [ $ u $ ] to produce some effect in an audience by means of the recognition of this intention”’.Footnote ² In keeping with this notion, the vast majority of research in semantics and pragmatics has focused on instances where the relation in (4c) holds (though see e.g. Franke, De Jager & Van Rooij Reference Franke, de Jager and van Rooij2012). As Campbell-Kibler (Reference Campbell-Kibler2008, Reference Campbell-Kibler2009), points out, however, very often in sociolinguistic research what a speaker’s utterance suggests to the hearer is not assumed to have been intended. Indeed, the remainder of this work makes clear that all of the relations in (4) are possible in practice, whether the source of a meaning is an utterance’s semantic content or its phonetic rendering.

2.3 Looking ahead

In Sections 3 and 4, I will employ the principle of intention-attribution in (3) and the relations enumerated in (4) to illuminate the nature of both meaning based in the indexical associations of variants of sociolinguistic variables, often referred to as social meaning, and meaning based in the semantic content of morphosyntactic expressions. Three major themes will emerge.

First, utterances can suggest information in diverse ways – some based in indexical associations, some based in semantics; some unintended, some overtly or covertly intended. Second, in accordance with previous sociolinguistic and pragmatic research, language users must contend with vast uncertainty in determining what exactly a speaker intends. Third, (3) furnishes principled insights concerning how language users navigate through that uncertainty and concerning the nature of the two types of meaning of interest.

Sections 3 and 4 focus primarily on indexically and semantically based meaning, respectively, but, for the sake of comparison, both sections will refer to both kinds of meaning at times. Section 3 is organized around the three intentionality relations in (4). Section 4 features two examples of debates about speakers’ intentions – debates which I show are rooted in the principle of intention-attribution in (3). In addition to elucidating the nature of meaning in practice in their own right, these two sections also set the stage for Section 5, where I show that game-theoretic models that only address meaning of the type in (4c) miss much of the crucial complexity and scope of meaning in practice.

3.1 No intention

This class of cases aligns with (4a), where something about a speaker $ S $ ’s utterance suggests something $ p $ , but $ S $ had no specific intention to suggest $ p $ . Examples abound in the sociolinguistics literature. Indeed, sociophonetic perception studies often find, as Campbell-Kibler (Reference Campbell-Kibler2008: 648) puts it, that ‘listeners feel entitled to read qualities into a speaker’s linguistic cues that speakers are unlikely to have included deliberately’. Levon (Reference Levon2014) and Tamminga (Reference Tamminga2017), for instance, report that speakers’ use of particular phonetic variants can suggest to hearers that the speakers are less ‘competent’ and ‘stupid[er]’. Coming off as stupid or incompetent is generally not something one intends to do (with some exceptions, no doubt), just as missing a shot on goal is generally something a footballer does not intend to do. Consistent with the desirability factor in (3a), then, these meanings are apt to being treated not as something the speaker intended to evince, but rather as revealing something about the speaker’s true nature.

Of course, unintended social meanings need not be undesirable. The main point here is that there are things our speech might signal that we have no specific intention of signaling, whether desirable or not. As just one example, if a speaker $ S $ , being from the United States, says something in $ S $ ’s native vernacular, $ S $ ’s accent may suggest that $ S $ is from the United States, even if suggesting as much was in no way a motivating factor for $ S $ ’s utterance or its phonetic shape.

3.1.1 Intention, $ L $ -alternatives, and meaning in language

This brings us to the first key theoretical point of this section, furnished by the $ L $ -alternatives factor of (3d). Example (3d) says an observer $ O $ is more likely to believe that an agent $ A $ intended to effect some consequence $ c $ by their action $ \alpha $ the more $ O $ thinks $ A $ believed that $ A $ was forgoing by selecting $ \alpha $ over $ L $ -alternatives – alternatives $ O $ believes $ A $ would think were accessible and less likely to effect $ c $ . Regarding the case of language, an important corollary of this principle is given in (5).

To illustrate, let us return to the case of a speaker $ S $ whose phonology suggests US nationality. Suppose that for all hearer $ H $ knows, $ S $ is using $ S $ ’s native phonology and for $ S $ to speak in a way substantially less likely to suggest US nationality would require greater effort and sound forced. In turn, from $ H $ ’s perspective any such alternative ways of speaking would be rather unattractive to $ S $ (assuming $ S $ does not want to sound pretentious), requiring more effort with no clear offsetting benefit. In this case, then, there’s rather little reason for $ H $ to think that $ S $ specifically intended to signal being from the United States phonetically. The situation is akin to the case of Pooh’s bees – as far as $ H $ knows, if $ S $ is to talk without expending unnecessary effort and sounding strange, $ S $ will evince US nationality simply by speaking. In other words, $ S $ is ‘just buzzing’.

Indeed, though not in exactly these terms, the matter of whether a speaker is ‘just buzzing’ – just speaking ‘naturally’ – is often found in discussions of social meaning and intentionality.Footnote ³ One thing that makes an action attractive for an agent is its being easy to commit, and speaking in a ‘natural’ manner (however real that notion actually is), intuitively means speaking less effortfully. In keeping with the discussion above, then, if a speaker $ S $ appears to be speaking naturally and hence relatively effortlessly, the evidence that $ S $ ’s phonetics were intended to serve some special purpose other than just articulating $ S $ ’s utterance diminishes. Contrariwise, if $ S $ appears to have forgone a more natural, easier way of talking, there is some reason to believe that $ S $ expended the extra effort toward some goal that $ S $ believes the more natural alternative would not have served as well.Footnote ⁴

This very dynamic shows up in Campbell-Kibler’s (Reference Campbell-Kibler2008: 648) analysis of how the speaker ‘Elizabeth’ is evaluated by participants in a study of perceptions of phonetic variants of the English -ing suffix: participants interpret Elizabeth as making a ‘sociolinguistic move’ precisely when she uses a variant that participants judge to be less ‘natural’ for her. The same basic reasoning about alternatives also helps explain Podesva’s (Reference Podesva2011: 237) claim that salient variants of sociolinguistic variables are particularly useful for conveying social meaning, where tokens are salient by being ‘infrequen[t]’ or ‘by exhibiting extreme acoustic values’. Insofar as there is a direct correlation between the frequency and the ease with which one produces a particular form, infrequent forms and forms exhibiting ‘extreme acoustic values’ are presumably more effortful than their counterparts. When a speaker uses such a form, then, there’s reason to believe that the speaker has forgone a less effortful option. If rational, the speaker would not exert extra effort for no reason, suggesting that the speaker aimed to achieve something that apparently otherwise attractive and less effortful options would be less likely to achieve. One plausible explanation, depending on the context, is that the speaker exerted extra effort in order to convey some social meaning.

Before moving on, it is worth noting that the principle in (5) operates not just vis-à-vis the socioindexicality of phonetic forms, but also relative to semantic content. (Indeed, being based in (3), it generalizes from utterances to actions more broadly.) The familiar phenomenon of scalar implicature (e.g. Horn Reference Horn2004) illustrates. Consider (6) and (7), both evaluated in an otherwise identical context where it is common ground that the respondent is sincere, fully knowledgeable about Terry’s performance, aware of terms like great, and prefers to be informative so long as it does not require lying.

The claim here is that though the respondent has given the same answer in both cases, (6) is more likely than (7) to indicate an intention to suggest that Terry’s performance was not great. This Gricean dynamic can be put in terms of the discussion in this section. With (6), the respondent had available alternatives less likely than satisfactory to suggest ‘not great’, like great. And, given what we know about the respondent, great is at least on some dimensions an attractive alternative, being more informative than satisfactory without requiring extra articulatory effort. The respondent’s choosing a response more likely to suggest ‘not great’ provides evidence for the possibility that the respondent wished to suggest as much and avoided otherwise apparently attractive alternatives like great to serve that goal.Footnote ⁵ With (7), however, there is no such alternative available, and the corresponding evidence vanishes.

To summarize this section, a corollary of the $ L $ -alternatives factor in intention-attribution in (3d), spelled out in (5), is that a speaker is more (less) likely to be taken to have intended to signal $ p $ by their utterance when it seems there was (was not) an attractive alternative less likely to suggest $ p $ that the speaker forwent. Concerning the phonetic rendering of one’s utterance, this means that if a speaker appears to be speaking in a manner in some way unusual for that speaker, the speaker’s phonetics are relatively likely to be taken as intended to serve some special purpose other than just articulating their utterance. If the speaker appears to be speaking ‘naturally’ and hence relatively effortlessly, however, the speaker’s phonetics are relatively unlikely to be taken as having some such special purpose.

3.1.2 Indexical meaning’s relative amenability to non-ascription of intention

The preceding discussion, especially regarding speaking ‘naturally’, foregrounds another point worthy of emphasis: meaning based in phonetic forms’ indexical (or iconic) character is relatively amenable to being perceived as being unintended.Footnote ⁶ This is because just as bees’ flying requires rapid vibration of their wings, speaking requires phonetics, which opens the question of whether the phonetic nature of the utterance was just buzzing – that is, whether it was simply in service of attempting to articulate the relevant morphosyntactic objects efficiently – or if it was indeed designed to convey something beyond the semantics of those morphosyntactic objects (and any pragmatic inferences derived therefrom).

The picture is rather different when it comes to the semantics of morphosyntactic objects. Again, as (5) indicates, believing that an aspect of an utterance was not intended to suggest something is facilitated by believing that there was no alternative available to the speaker given their other goals and constraints. But words and phrases typically do have accessible alternatives, often including saying nothing at all. The absurd exchange in (8) illustrates.

Bees’ buzzing is incidental to their flying, and my variants of /s/ and /i/ in a given context may simply be in service of articulating the word see intelligibly and relatively effortlessly, but it would take very special circumstances for one to observe A’s first utterance in (8) and conclude that A’s use of the words mug, under, and green had nothing to do with evoking their semantic content. Why should A go to the trouble of saying those particular words if A meant nothing by them, especially as part of an ostensibly communicative act? It is indeed generally very strange (though see Section 4.1) to issue a morphosyntactic object, with all of its encoded semantic meaning, with no intention of suggesting something involving that meaning. Morphosyntactic objects, then, are generally harder to write off as just buzzing. Rather, they are generally taken to be intended to suggest something in their own right.

The notion that indexical meaning is relatively amenable to being read as not intended figures importantly in cases of indexical meaning bearing the relation in (4b), to which I now turn.

3.2 Intention with intention for nonrecognition of intention

This class of cases aligns with (4b), where the speaker intends to suggest something and for that intention to go unrecognized. Such cases in a sense depend on the class of cases above: the speaker hopes that the hearer will misidentify the relevant meaning as unintended. I may, for instance, want to sound smart, cool, etc., but emphatically not want you to recognize as much, as recognition may undermine my intention. Campbell-Kibler (Reference Campbell-Kibler2009: 137) puts it this way: ‘[Some meanings,] such as jaded, lose their meaning when recognized as intentional’. This dynamic is illustrated clearly in Campbell-Kibler’s (Reference Campbell-Kibler2008: 655) analysis of participants’ evaluation of speaker ‘Valerie’, whose ‘use of -ing [as opposed to -in’]… successfully means intelligent to exactly those listeners who do not perceive that move as intentional. Those who think she intends “intelligence” by her social cues react by seeing her as less intelligent’.

Thus, successfully suggesting certain things sometimes requires that the relevant intention go unrecognized. And this stands to reason. If we are trying to discern the kind of person someone is or the state they are in, we want to get it right, and the appearance of authenticity can be crucial. Being perceived as authentic often depends on being perceived as not going out of one’s way to convince others of something about oneself – that is, it often depends on the observer’s believing that one is just letting one’s true self or state show. A child’s cough, for instance, will only earn them a day home from school if it is believably unintentional.

As just established in Section 3.1.2, meaning based in the indexicality of phonetic forms is particularly amenable to being perceived as unintended. In turn, it is likewise relatively useful for meanings that speakers intend, but want to be perceived as unintended.

3.3 Intention with intention for intention-recognition

This class of cases lines up with (4c). Podesva’s (Reference Podesva2011) research on high-rising terminals (HRT) provides an example. Analyzing previous research and his own data on HRT, Podesva (Reference Podesva2011: 245) suggests that across various uses, HRT ‘serves as a politeness strategy enabling the speaker to express concern for the hearer’. Expressing concern for one’s hearer is generally compatible with intending to have one’s intention be recognized. King’s (Reference King2018) research on language in a sex-education class, where a student selectively employs ‘Hip Hop styling’ prosody to overtly index dominance, provides another example of using phonetics to convey social meaning and intending to have that intention recognized.

One can find similar dynamics at the segmental level. Acton (Reference Acton, Hall-Lew, Moore and Podesva2021) discusses a case involving a speaker $ S $ who is a highly educated, friendly, scarce user of the -in’ form of the -ing suffix, where this information about $ S $ is common ground between $ S $ and the hearer $ H $ . Asked about $ S $ ’s weekend plans, $ S $ says, ‘I’m goin’ fishin’!’. Given that $ S $ ’s phonetic tendencies are common ground between $ S $ and $ H $ here, it is easy to imagine that $ S $ used -in’ instead of -ing hoping that $ H $ would recognize that $ S $ used what for $ S $ is a rare and thus presumably relatively costly form in order to suggest something. ‘If $ S $ didn’t intend to suggest anything by their phonetics’, $ S $ might count on $ H $ to ask, ‘why didn’t $ S $ use the “cheaper” form that we both know $ S $ uses more frequently?’ In that case, $ S $ both attempts to signal something and intends for that attempt to be recognized.

$ H $ ’s conclusions about what exactly $ S $ was trying to signal would depend on multiple factors. As Acton (Reference Acton, Hall-Lew, Moore and Podesva2021) notes, one possible conclusion would be that $ S $ invoked the trait-based meanings of -in’ (e.g. suggesting the speaker is easygoing) not to say something general about $ S, $ but to say something more local – for instance, that the fishing event will be a casual affair.Footnote ⁷ In any case, what is important here is that $ S $ may well intend to suggest something to $ H $ and have $ H $ recognize that intention.

Before turning to the role of intention in meaning at the morphosyntactic level, I wish to highlight the general importance of common ground in cases where the speaker intends for their communicative intention to be recognized. As has been referenced several times, (5) suggests that a hearer $ H $ ’s believing that a speaker $ S $ intended to suggest something $ p $ is facilitated by $ H $ ’s believing that $ S $ forwent an attractive $ L $ -alternative. The related claim here is that $ H $ ’s believing that $ S $ intended for $ S $ ’s intention to suggest $ p $ to be recognized is facilitated by it being common ground (in the sense of e.g. Stalnaker Reference Stalnaker2002) between $ S $ and $ H $ that $ S $ indeed had access to an attractive $ L $ -alternative.

The goin’ fishin’ example provides an illustration of this very general dynamic. Here, $ S $ relies on $ H $ ’s knowledge that, for $ S $ , -ing is an attractive $ L $ -alternative to -in’, being $ S $ ’s more frequent variant. Because of $ H $ ’s knowledge, $ S $ can count on $ H $ to understand that by using -in’, $ S $ is ‘going out of the way’ phonetically, potentially for some special purpose. In contrast, if the scenario were such that $ S $ believed $ H $ had no knowledge of $ S $ ’s phonetic tendencies, $ S $ would in turn have less reason to expect $ H $ to conclude that $ S $ was going out of the way phonetically to communicate something special by using -in’ rather than -ing.

In this way, one’s knowledge (or lack thereof) of speakers’ repertoire and tendencies – that is, what options are available to them and the relative costs of production of each – has a crucial impact on one’s interpretation of their utterances. It makes good sense, then, that studies such as Podesva (Reference Podesva2011) and Podesva et al. (Reference Podesva, Reynolds, Callier and Baptiste2015) offer speaker-specific analyses of the social meanings of phonetic forms that incorporate facts about individual speakers’ usage patterns. Very generally, such facts play a pivotal role in discerning what speakers intend and, in turn, what we take away from their utterances.

3.4 Summarizing

In this section, we have seen that meaning based on the indexicality of variants of linguistic variables can occur with any of the three intention relations in (4). A phonetic aspect of a speaker’s utterance may suggest something despite the speaker having no intention for it to do so, or the speaker may intend for it to suggest something hoping that that intention will not be recognized or that it will. Nor is it surprising that the first two classes of intention relations receive so much attention in sociolinguistic research. For one, though it is often unclear that a speaker meant anything by the phonetic character of their utterance, that phonetic character is virtually always potentially meaningful in the broad sense of suggesting something to the hearer. Secondly, signaling certain things depends on not appearing to be trying to do so (Section 3.2.1).

I also showed in this section that, though speakers’ sociophonetic intentions cannot be known with certainty, (3) provides principled insights into how language users attempt to discern such intentions (or lack thereof). Playing a starring role in that regard was (5) – a corollary of the $ L $ -alternatives factor in (3d) – which states that believing a speaker intended to signal something $ p $ via their utterance is facilitated by believing the speaker forwent an attractive $ L $ -alternative. (I further showed that a hearer $ H $ ’s believing that the speaker $ S $ intended for their intention to be recognized is facilitated by it being common ground between $ S $ and $ H $ that $ S $ forwent such an alternative (Section 3.3).) Example (5) squares with the findings of previous sociophonetic perception studies, helps derive Podesva’s (Reference Podesva2011) claim that ‘salient’ variants of linguistic variables are particularly useful for conveying social meaning, and explains why, relative to meaning based in the semantics of morphosyntactic objects, meaning based in the indexical character of phonetic forms is particularly amenable to being perceived as unintended.

On this last point, though meaning based in the semantics of morphosyntactic expressions may generally be less amenable to being perceived as having no communicative intention behind it at all, it is not entirely unamenable to such perceptions. Moreover, what exactly a speaker intends to convey via their morphosyntax is very often open to question. Thus, as I will show in the next section, meaning of this type – like indexically based meaning – is profoundly underdetermined, and determining what speakers intend by the semantics of their utterances is governed by the same principles of intention-attribution in (3) applied in the foregoing discussion.

4.1 Morphosyntactic objects without speaker intention

There are certainly some cases where individuals issue a word or more without themselves intending to suggest anything by it. Involuntary speech (e.g. sleep-talking) presumably falls under this category. Verbal ‘tics’ may fit under this rubric in some cases, too, though it should be noted that research on filled pauses, for instance, has revealed correlations between particular filled pauses (e.g. um vs. uh) and particular discourse situations (e.g. Clark & Fox Tree Reference Clark and Fox Tree2002), suggesting that the forms are, at minimum, differentially suited for particular intentional purposes. Then there are cases where one speaks on another’s behalf (rather than speaking to communicate something one independently wishes to convey), which Levinson (Reference Levinson, Drew and Wootton1988) argues have received insufficient attention. Du Bois (Reference Du Bois, Hill and Irvine1993) and Gaenszle (Reference Gaenszle2016), for instance, discuss the case of divination, where the speaker may simply be a channel for a deity’s message.

Expressives, too, present an interesting case. On some analyses, felicitously using expressives does not strictly require an intention to suggest anything (see e.g. Bach Reference Bach2006), though it may reveal something about a speaker’s emotional state. It certainly seems that we sometimes utter expressives not to suggest anything to anyone but for catharsis (Wharton Reference Wharton2016) – one may exclaim, ‘Damn!’ upon learning bad news with no one else around, not to communicate but to release anger. But as Blakemore (Reference Blakemore2015) and Wharton (Reference Wharton2016) point out, expressives can also be used with an intention to suggest something, even ostensively – for example, to make clear to others one’s emotional reaction to something.

Aside from involuntary/compulsory or cathartic speech and channeling the speech of others, however, it is rather rare for one to speak and be perceived as having no intention to suggest anything by the semantics of one’s utterance. The remainder of this section concerns cases where a speaker does indeed appear to intend something by their utterance, but what exactly was intended is the subject of debate. While cases where speakers intend to have their intentions recognized are the bread and butter of semantic and pragmatic research (though see e.g. Franke et al. Reference Franke, de Jager and van Rooij2012), in practice there is often uncertainty even at the morphosyntactic level about what a speaker intends to suggest. The following examples illustrate how reasoning about what in particular a speaker intended to suggest is grounded in the theory of intention-attribution outlined in (3).

4.2 Example: Ilhan Omar’s statements on Israel and its supporters

In this example it is clear that the speaker meant something by her utterances (and something negative at that), but there was controversy concerning whether she specifically intended for her words to have anti-Semitic force. The case involves US congresswoman Ilhan Omar, who in 2018 and 2019 drew considerable attention for statements she made criticizing Israel and its supporters. The most controversial statements are provided here in (9). Space limitations preclude a comprehensive treatment of these statements and their contexts, nor do I intend to endorse a particular interpretation of them. My focus here is on how discussions of Omar’s intentions center on the principle in (3).

Many commentators pointed out that Omar’s statements in (9) bear connections to anti-Semitic stereotypes (e.g. Beauchamp Reference Beauchamp2019; Weiss Reference Weiss2019), and there was no shortage of speculation concerning whether Omar intended for her statements to bear anti-Semitic force. ‘Omar, I suspect, knows exactly what she is doing’, wrote New York Times columnist Bret Stephens in a March 7, 2019, opinion piece on Omar’s statements, and many readers commented for or against this claim (Stephens Reference Stephens2019: n.p.). A few illustrative examples are provided in (10).

Example (10a) suggests that it is not credible that Omar invoked three different anti-Semitic tropes simply by chance and hence must have known her words would have anti-Semitic force. This argument links to the ‘expectation’ factor of (3b), which says that we are more likely to think someone intended an outcome $ c $ the more we think they expected their action to effect that $ c $ . In terms of our metaphor from Section 2.1, (10a) and Stephens suggest that Omar’s actions are not akin to an unwitting two-year-old pulling a fire alarm. Example (10b) gives a similar assessment, suggesting that, at best, Omar has been indefensibly careless in wording her criticisms of Israel. Example (10c), alternatively, offers a counterargument to (10a–b), appealing to the $ L $ -alternatives factor in (3d) (though not in those words): according to (10c), there are no alternative utterances critical of Israel and its supporters but less likely than Omar’s statements to be linked to anti-Semitic tropes. In other words, according to (10c), there is simply no way for Omar to criticize Israel without being accused of evoking anti-Semitic tropes, even if the latter is not among her goals.

Example (10d) defends Omar against claims of anti-Semitism by suggesting that if she had wanted to criticize Jewish or Israeli people in general she could have done so explicitly, rather than talking about Israel as a state. Linked to the $ M $ -alternatives factor in (3c), the reasoning here is that among the alternatives accessible to Omar were utterances more likely to have clear anti-Semitic force, but Omar opted against such alternatives, suggesting that expressing anti-Semitism was not her intention. In terms of our earlier metaphor for this factor, (10d) is in effect saying that we have a situation like a speaker saying damn among friends: if the speaker really wanted to express extreme displeasure, they could have used a more taboo expletive. Conversely, the $ M $ -alternatives factor in (3c) also provides a basis for a counterargument because it takes into account the costs of alternatives. So, while there were accessible alternatives more likely to express unequivocal anti-Semitism, such alternatives may have been so socially costly – perhaps costing Omar her seat in Congress – that, on balance, they offered Omar relatively little even if she did wish to signal anti-Semitism. Looking beyond this particular case, (3c) is what permits one to think that although a speaker forwent more direct or potent routes to effecting some outcome $ c $ , they may still have intended to effect $ c $ , avoiding the alternatives simply because they were too costly. This recalls the case of the speaker saying damn while teaching a class of kindergartners – the speaker may have forgone a stronger expletive not for lack of intense displeasure but because stronger language might have been more likely to lead to censure.

Stephens’s headline and the comments in (10) thus all appeal to the factors in (3), making direct or indirect connections to what Omar might reasonably have expected her utterances to convey or to the properties of alternative utterances more or less likely to convey anti-Semitism.

4.3 On virtue signaling

There is no shortage of other examples where a speaker’s intentions in issuing some semantic content are underdetermined by that content and potentially covert. In discussing social meaning, I mentioned that speakers may use specific phonetics intending to suggest something about themselves, hoping for that intention to go unrecognized – for example, to sound cool or intelligent. The same goes for semantic meaning. Bach (Reference Bach, Allan and Jaszczolt2012: 52), for instance, notes that one might say self-deprecating things in order to appear modest, hoping for that intention to go unrecognized because recognition of the intention ‘may vitiate it’. Such is the nature of so-called ‘virtue signaling’, where one expresses or does something supposedly virtuous primarily to appear virtuous. Accusations of virtue signaling abound on the Web, as in the tweet in (11) from August 4, 2019. Here the user, whose other tweets support gun rights and the abolition of government, comments on reactions to two mass shootings in the United States from the day before.Footnote ⁹

The user argues that it would be a waste of energy for the individuals in question to condemn the shootings simply to inform the public about their stance on the shootings, since it would be taken for granted that they were against the shootings. Hence, according to the user, there must have been some extra benefit to condemning the shootings: namely, appearing virtuous. Put in terms of the $ L $ -alternatives factor in (3d), according to this user, the alternative of saying nothing, which would be less likely to look like an attempt to appear virtuous, still has much to offer, requiring less effort and being consistent with being against the attacks (which would supposedly be taken for granted). Therefore, the argument goes, there is good reason to think that individuals in question were motivated by a desire to display virtuousness. This argument is not unassailable, however, as users responding to the tweet indicate. Multiple critiques argue that there are reasons for explicitly condemning the shootings other than attempting to appear virtuous, even if one’s condemnation may be taken for granted. User @phoebesux6, for instance, writes, ‘So expressing any feelings about people being dead is?? Virtue signaling?’

4.4 Summary

While it’s rare for a speaker not to intend to suggest anything at all by the semantics of their utterance, what exactly the speaker does intend to suggest is not always clear, and can be both contentious and consequential. And the same principles that govern intention-attribution with respect to sociophonetic meaning, which center on beliefs about the agent’s preferences, knowledge state, and appraisal of alternatives, apply just as well here. The preceding discussion thus echoes Franke et al.’s (Reference Franke, de Jager and van Rooij2012) call for increased attention in pragmatic research to scenarios where the usual assumptions of Gricean cooperativity and overt intention are not taken for granted.

Continuing along these lines, in the next and penultimate section of this paper, I turn to the role of intention in Burnett’s (Reference Burnett2017, Reference Burnett2019) social meaning games models (SMGs) and the rational speech act models (RSAs) (e.g. Goodman & Frank Reference Goodman and Frank2016) to which they are related. These models have greatly enriched our understanding of how messages are conveyed in linguistic exchanges. At the same time, I will explain why the foregoing discussion points to a need to make the models more complex. For example, as discussed in Section 3 and contrary to what SMGs and RSAs are designed to account for at present, speakers sometimes hope their intentions will go unrecognized, certain social meanings depend on being performed in a way that seems natural, and hearers may conclude things from utterances that they think the speaker had no intention of suggesting.

5.1 Background

Central to game-theoretic models of language use and interpretation (e.g. Franke Reference Franke2009; Frank & Goodman Reference Frank and Goodman2012; Burnett Reference Burnett2017) is the idea that speakers and hearers reason about each other’s beliefs in choosing and interpreting utterances and that they know this about each other. A speaker $ S $ wishing to convey information to a hearer $ H $ attempts to select the utterance that appears to offer the best mix of being inexpensive and likely to convey exactly the information $ S $ wishes to convey. Similarly, observing some utterance $ u $ , $ H $ interprets $ u $ relative to $ H $ ’s prior beliefs and the assumption that $ S $ selected $ u $ as just described. The interaction can be construed as a game in that there are agents employing strategy to achieve some goals. Influential in such game-theoretic research are RSA models, which situate language use and interpretation in broader theories of cognition as probabilistic reasoning (Goodman & Frank Reference Goodman and Frank2016). RSA models have been applied to wide-ranging phenomena, offering means for testing quantitative predictions about language use and interpretation and delivering general insights into how these processes work (see Goodman & Frank Reference Goodman and Frank2016 for multiple examples).

Building on the insights and architecture of RSAs (as well as Franke Reference Franke2009), Burnett (Reference Burnett2017, Reference Burnett2019) broadened the scope of game-theoretic research on meaning to include social meaning via her SMG models. In SMGs, a speaker $ S $ , choosing which variant of a linguistic variable to use, tries to select the best variant for projecting the persona $ S $ wishes to project to the hearer $ H $ . $ H, $ in turn, tries to figure out what $ S $ is trying to convey – both parties taking into account $ H $ ’s prior beliefs about $ S $ ’s persona. The principal difference between RSAs and SMGs is that whereas RSAs typically involve signals/inferences concerning some question under discussion (Roberts Reference Roberts1996) based on an utterance’s semantic content, SMGs involve signaling/inferring something about $ S $ ’s desired persona based on the indexical field (Eckert Reference Eckert2008) of the variant $ S $ uses.

5.2 Cooperativity, reliable information, and presumption of intention

Among the shared assumptions for RSAs and SMGs to date is a version of Gricean (Grice Reference Grice, Cole and Morgan1975) cooperativity. In these models, it is common ground among $ S $ and $ H $ that $ S $ ’s goal in issuing a particular utterance is to provide as much relevant and reliable information to $ H $ as possible, modulo considerations of cost, and that $ H $ ’s goal is to infer the information $ S $ is attempting to convey and update their beliefs with that information.Footnote ¹⁰ In these models, then, both parties’ interests are in effect the same as regards the utterance under consideration. As Burnett (Reference Burnett2019: 11) puts it: ‘[B]oth players win if [ $ H $ ] correctly interprets $ S $ ’s message, updating their beliefs accordingly, and they both lose if… [ $ H $ ] comes to believe something different about the world than that which $ S $ intended’.

Two points here merit attention. The first point is simply that these models are designed to capture only meanings that $ H $ believes $ S $ intended to signal. The second concerns the assumption that $ S $ will only attempt to convey reliable information. In SMGs, where personae and indexical fields are modeled as sets of properties, this assumption is operationalized by assuming that $ H $ rules out any potential persona for $ S $ that shares no properties with the indexical field of the variant $ S $ uses. For instance, modeling the indexical fields of -ing and -in’ as {competent, delicate} and {incompetent, casual}, respectively, Burnett (Reference Burnett2017: 258) notes that in SMGs, if $ H $ hears ‘-ing, they discard the possibility that the speaker is [both] incompetent and casual’.

The formal nature of SMGs has sharpened our understanding of the dynamics of social meaning by requiring explicitness about how things are supposed to work and the assumptions upon which the system rests. As may already be apparent, however, SMGs to date do not fully square with empirical observations about social meaning in other literature. In the next subsection, I will point out some of the challenges SMGs face, drawing on the preceding discussion. I will then show that RSAs to date face related challenges, made manifest in the light of SMGs’ challenges, lessons from sociophonetic research, and the discussion herein.

5.3 Challenges for SMGs

A key set of difficulties for SMGs stems from the assumptions that $ S $ and $ H $ mutually hope $ H $ will correctly apprehend $ S $ ’s intended message and update their beliefs accordingly, and that $ H $ assumes $ S $ ’s selected variants always provide reliable information about $ S $ ’s personae. It is certainly not always in $ H $ ’s best interest to believe whatever $ S $ intends for $ H $ to believe. Rather, $ H $ would more generally be well served by trying to figure out what $ S $ is actually like. Whether $ S $ is truly competent or friendly or just attempting to appear so, for instance, can have important consequences for $ H $ .

Indeed, the discussion in Section 3.2.1 suggests that, ceteris paribus, differences in the nature of sociophonetic meaning and semantic content provide a reason for hearers to be quicker to doubt the reliability of an apparently intended sociophonetic meaning than the reliability of an utterance’s semantic content. Franke et al. (Reference Franke, de Jager and van Rooij2012) and McCready (Reference McCready2015) point out that if a speaker is found to willingly provide unreliable information, it can cost that speaker in terms of social capital (as in the case of the boy who cried, ‘Wolf!’). Even amoral speakers thus have some incentive to tell the truth in the long run, particularly regarding easily verifiable matters of fact (though other considerations may trump that incentive). However, as also discussed in Section 3.2.1, given the performative and (inter)subjective nature of sociophonetic social meaning and its relative amenability to denials of intention by the speaker, it is relatively difficult to establish conclusively that a speaker suggested unreliable information via the phonetics of their utterances, not to mention that they did so deliberately. Therefore, we might generally expect the extrinsic incentive for speakers to provide reliable information to be weaker with sociophonetic meaning than with, say, the semantics of objective assertions – creating room for doubt about the reliability of what a speaker’s phonetics might indicate about the speaker.

Moreover, predictions aside, the empirical research discussed in Section 3 has clearly shown that hearers do not always interpret a speaker’s phonetics in the way they think the speaker intended.Footnote ¹¹ Recall Campbell-Kibler’s (Reference Campbell-Kibler2008: 655) case of ‘Valerie’: ‘[Participants who] think she intends “intelligence” by her social cues react by seeing her as less intelligent’. SMGs in their current form have no way of accounting for such cases. Given the assumption of cooperativity as implemented in SMGs, $ H $ assumes that $ S $ will provide reliable information, and if $ H $ thinks that $ S $ intends to signal some property via $ S $ ’s chosen variants, $ H $ will ascribe that property to $ S $ .

Podesva et al.’s (Reference Podesva, Reynolds, Callier and Baptiste2015) study of perceptions of US politicians poses similar problems for SMGs. For instance, the authors found that Barack Obama’s speech was ‘rated as sounding more intelligent’ when his word-final /t/s were unreleased than when they were released. Research on word-final /t/ in English suggests that it is the released variant that would be more likely to be associated with intelligence, given its well-documented associations with traits like articulateness and learnedness (Benor Reference Benor2001; Bucholtz Reference Bucholtz2001). But, again, given the assumption of SMGs that hearers will take speakers at their word, so to speak, we should be surprised to find a speaker being perceived as having a lower degree of some property (like intelligence) when they use a variant more closely associated with that property (like released /t/) than when they use another.

More generally, as discussed extensively in Section 3, that a hearer $ H $ believes that a speaker $ S $ intends to convey $ p $ via the phonetics of their utterance is neither necessary nor sufficient for $ H $ coming to believe $ p $ on the basis of $ S $ ’s phonetics. $ H $ may infer things that $ S $ had no intention of suggesting; $ H $ may reject some meaning $ H $ thinks $ S $ intended to convey; and so on. Nor are such cases marginal in the realm of social meaning. But given SMGs’ assumptions about intentionality and cooperativity, they fall outside of SMGs’ scope in their present form.

5.4 Summarizing, and looking ahead

That said, though such cases of sociophonetic meaning are not accounted for by SMGs at present, they are presumably governed by many of the same general principles of probabilistic reasoning, social recursion, and strategic use and interpretation of linguistic resources that SMGs currently employ. It is just that such cases do not involve such strong assumptions about speakers and hearers’ intentionality, common ground, and shared objectives. Rather, such cases call for a more complex model in which, with respect to a particular sociophonetic variant $ v $ , $ S $ might well be uncertain about, for instance: (i) whether $ H $ would interpret $ v $ as suggesting something about who $ S $ is; and (ii) whether $ H $ would interpret $ v $ as being motivated by a desire to suggest something about who $ S $ is. On the flipside, $ H $ may well be uncertain about: (i) whether $ S $ intended to suggest anything via $ v $ ; and (ii) how well the indexical field of $ v $ squares with $ S $ ’s actual personality, stances, etc. Adding complexity along these or similar lines is necessary to cover the complex reality of social meaning. Franke et al.’s (Reference Franke, de Jager and van Rooij2012) work on game-theoretic pragmatics without the strong assumption of cooperativity could prove helpful in this regard (see also Franke Reference Franke2013). While that work focuses on inferences based in semantics, it proceeds from the fact that speakers and hearers need not have the same interests but may still apply strategy and reason in using and interpreting language. It is my hope that the principles developed herein concerning how speakers navigate uncertainty about intention will likewise help shape models of social meaning with broader scope.

Before concluding, I turn to a couple of related challenges for RSAs to date – challenges brought into relief by the foregoing discussion and insights from sociophonetic research.

5.5 Related challenges for RSA models

Campbell-Kibler (Reference Campbell-Kibler2008: 654) rightly points out that social meaning in situated use does not simply ‘reside in the speaker’s intention’, thus distinguishing it from the type of meaning with which Grice (Reference Grice1957) was fundamentally concerned. But it is worth further pointing out that, in practice, meaning in context never resides entirely in the speaker’s intention, whether it is based in semantics, indexicality, or something else. Franke et al. (Reference Franke, de Jager and van Rooij2012), for instance, examine a range of a cases in which a rational hearer might infer something from an utterance’s semantics that the speaker did not intend at all.

Indeed, lessons from sociophonetic research about gaps between intention and interpretation point to analogues within semantically based meaning that present challenges for models that assume cooperativity. Take for instance the phenomenon of ‘hypercorrection’ (Labov [1966] Reference Labov2006), whereby a speaker, apparently attempting to signal prestige, uses a prestigious variant of a variable more than would members of a more prestigious group with which the variant is associated. By overdoing it, speakers may end up being perceived as not having the relevant property (in this case, prestige). As noted above, SMGs cannot at present account for such cases; they assume hearers will regard speakers’ sociolinguistic performances as ‘truthful’ regardless of how convincing they really are.

This same dynamic can occur with semantically based meaning. Just as one can overuse a sociophonetic variant, one can assert something so many times or with such force as to lead the hearer to suspect that the assertion is false. We see this in Queen Gertrude’s reaction to the play within the play in Hamlet, in which her counterpart ‘protest[s] too much’ (Shakespeare Reference Shakespeare1998: act III, scene ii): the play’s queen’s vows of faithfulness to her husband are so overdone as to seem insincere. Like SMGs, RSAs are not presently designed for cases in which a speaker intends to communicate $ p $ repeatedly, and the hearer recognizes as much, but, eventually, on the basis of the speaker’s repeated assertions, concludes not $ p $ .

More generally, RSAs are not presently designed for cases where the hearer believes that the speaker intends to communicate $ p $ but concludes not $ p $ . Kao & Goodman (Reference Kao and Goodman2015) provide an insightful account of irony, which includes hearers concluding that the opposite of the semantic content of the speaker’s utterance is true. But in the cases their models account for, the hearer assumes that speaker intended for the hearer to come to that conclusion – that is, that the speaker wanted the hearer to conclude not $ p $ even though the semantics of their utterance was $ p $ – so the assumption of cooperativity remains.

Just as ‘protesting too much’ can raise suspicions about one’s sincerity, even a single instance of an utterance or variant can be delivered in such a way that hearers take the opposite of an apparently intended meaning to be true. I have already discussed sociophonetic examples of this. But achieving one’s goals via the semantics of one’s utterance often requires a credible performance, too. Indeed – and this is crucial – all speaking is a performance of sorts, and using aspects of our utterances to change others’ belief states in the way we intend requires getting our performances right.

Discourse surrounding a May 2019 New York Times report that, from 1985 to 1994, then private citizen Donald Trump lost more than a billion dollars (Buettner & Craig Reference Buettner and Craig2019) provides an instructive case. When the hosts of the Fox News morning show Fox & Friends discussed the story, they depicted it in a generally positive light. Co-host Ainsely Earhardt said of the story:Footnote ¹²

Many observers questioned whether Earhardt could sincerely believe that losing so much money is impressive (in a positive sense), and some suggested her delivery betrayed that she didn’t actually believe her words. Aspects of her delivery drawing commentary included the focus of her gaze and the fact that she raised her hand in front of her mouth on the phrase most of us could ever achieve, taking an arrhythmic pause before the word achieve. Example (13) presents a handful of reactions from both media personalities and casual observers.

Each reaction suggests that something about Earhardt’s delivery indicates that she does not truly believe her words. But the comments make no suggestion that Earhardt wants to come across that way. Rather, the interpretation seems to be that she intends to convey that she does believe what she is saying, but fails to do so. This is in contrast to cases of (cooperative) irony, where the speaker intends to suggest the opposite of what they say and wants the hearer to understand that.

Thus, just as appealing to the indexicality of a phonetic variant does not guarantee that one’s intended effect will be achieved, so it goes with the semantics of one’s utterances. Successfully making an explicit public commitment to a belief or preference (Condoravdi & Lauer Reference Condoravdi and Lauer2012) means selling it, and sometimes further requires selling it without looking like you are trying to do so. Even expressives, which Potts (Reference Potts2007: 167) argues ‘do not offer content so much as inflict it’, can be unconvincingly performed. While it is true that the content of things like swear words is not open to the same mechanisms of denial as, say, at-issue content, a hearer will only believe a speaker is in the emotional state conventionally associated with the relevant word if the delivery of that word is right. Again, while we may generally default to taking speakers at their word (and as Franke et al. Reference Franke, de Jager and van Rooij2012 point out, for good reasons), carrying off one’s intended communicative goal means delivering a credibly sincere performance. This may especially be the case for expressions of emotion, opinion, or evaluation like (12), where, as with the case of social meaning (see Section 3.2.1), much of what is suggested cannot be independently verified by objective means.

In brief, though mismatches between what is inferred and what is intended may be more frequent or readily apparent with social meaning, semantics-based meaning is subject to the same dynamics. Developing a game-theoretic model of language use that can accurately predict inferences in cooperative contexts and also detect when utterances intended to be viewed as sincere are likely to be considered insincere is a tall order (see e.g. Franke et al. Reference Franke, de Jager and van Rooij2012; Kao & Goodman Reference Kao and Goodman2015 for some steps in that direction). But humans’ use and interpretation of language demonstrably involve both, among other complicating factors. Not only do we sometimes doubt the truth of an utterance’s semantics, but, relatedly, the cliché that it is not just what you say but how you say it is true, and these are facts that we as speakers and hearers must contend with every day. Ultimately, so, too, must our theories of language use and interpretation.